耦合台风天气预报模式和实测数据的神经网络风速预测  

Neural network forecasts of typhoon wind speeds coupled with WRF and measured data

作　　者：黄铭枫[1,2] 刘国星 王义凡 徐卿 HUANG Mingfeng;LIU Guoxing;WANG Yifan;XU Qing(Institute of Structural Engineering,Zhejiang University,Hangzhou 310058,China;Center for Balance Architecture,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]浙江大学建筑工程学院,浙江杭州310058 [2]浙江大学平衡建筑研究中心,浙江杭州310058

出　　处：《建筑结构学报》2022年第3期98-108,共11页Journal of Building Structures

基　　金：国家自然科学基金项目(51838012,52178512);浙江省自然科学基金资助项目(LZ22E080006)。

摘　　要：基于准定常假定,风荷载与风速平方成正比。为了实现对结构的台风动力效应进行分析预测,建立了耦合数值天气预报(weather research and forecast,WRF)模式和现场实测数据的风速预测神经网络模型,开展台风短期风速的高精度预测。利用该模型对2017年"泰利"和2018年"康妮"的台风风场进行模拟和预测。为了提高神经网络训练效率和短期风速预测精度,使用经验模态分解和小波变换对风速数据进行分解,将分解后的风速分量作为神经网络的输入。结合数据分解方法,分别采用BP、Elman、GRNN和ANFIS等4种神经网络模型完成了两次台风影响下某实测场地平均风速的6步提前预测。结果表明,基于集合经验模态分解或小波分解的神经网络预测方法,相比不进行数据分解的直接神经网络预测法,台风风速预测精度提高了50%以上。在4种神经网络模型中,小波变换和BP神经网络模型组合具有最优的台风风速预测精度。Based on the quasi-constant assumption,the wind load is proportional to the square of the wind speed.In order to realize the analysis and prediction of the typhoon dynamic effects of the structure,artificial neural networks coupled with weather research and forecast(WRF)and measured data were developed and trained to forecast the short-term wind speed of Typhoon Talim in 2017 and Typhoon Kong-rey in 2018.The wind speed data were decomposed by wavelet(WT)decomposition and ensemble empirical mode decomposition(EMMD)to improve the training efficiency of the neural network and the accuracy of short-term wind speed prediction.The obtained components would be served as inputs for the neural network.Combined with the data decomposition method,four kinds of neural network models,i.e.,back propagation(BP),Elman,general regression neural network(GRNN),and adaptive neural fuzzy inference system(ANFIS)were employed to carry out a six-step prediction of the mean wind speed at the full-scale measurement site under influences of Typhoon Talim and Kong-rey.The result shows the EMMD or WT-based neural networks improve the prediction accuracy by more than 50%compared to the direct neural network method without data decomposition.It is found that the combination of wavelet decomposition and BP neural network can achieve the best prediction accuracy of typhoon wind speed.

关 键 词：台风风速预测 神经网络 数据分解 天气预报模式 

分 类 号：TU312.1[建筑科学—结构工程] TU311.41